From nobody Sun Feb 8 18:31:33 2026 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 12FF7EB64D9 for ; Tue, 11 Jul 2023 03:05:34 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231414AbjGKDFb (ORCPT ); Mon, 10 Jul 2023 23:05:31 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:48784 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231195AbjGKDE7 (ORCPT ); Mon, 10 Jul 2023 23:04:59 -0400 Received: from szxga08-in.huawei.com (szxga08-in.huawei.com [45.249.212.255]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 0EACC10EF for ; Mon, 10 Jul 2023 20:03:45 -0700 (PDT) Received: from dggpeml500022.china.huawei.com (unknown [172.30.72.56]) by szxga08-in.huawei.com (SkyGuard) with ESMTP id 4R0Qfk6hSjz1FDn8; Tue, 11 Jul 2023 11:03:10 +0800 (CST) Received: from huawei.com (10.175.112.208) by dggpeml500022.china.huawei.com (7.185.36.66) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.27; Tue, 11 Jul 2023 11:03:43 +0800 From: Zhang Zekun To: , , , , CC: , , , , Subject: [PATCH -next 1/2] iommu/iova: Add check for cpu_rcache in free_iova_rcaches Date: Tue, 11 Jul 2023 10:54:51 +0800 Message-ID: <20230711025452.35475-2-zhangzekun11@huawei.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20230711025452.35475-1-zhangzekun11@huawei.com> References: <20230711025452.35475-1-zhangzekun11@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.175.112.208] X-ClientProxiedBy: dggems701-chm.china.huawei.com (10.3.19.178) To dggpeml500022.china.huawei.com (7.185.36.66) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" free_iova_rcaches() needs to check if cpu_rcache->loaded and cpu_rcache->prev is NULL before freeing them. Because iova_domain_init_rcaches() may fail to alloc magazine for cpu_rcache->loaded and cpu_rcache->prev, but they will be freed for all cpus. Fixes: 32e92d9f6f87 ("iommu/iova: Separate out rcache init") Signed-off-by: Zhang Zekun --- drivers/iommu/iova.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c index 10b964600948..3c784a28e9ed 100644 --- a/drivers/iommu/iova.c +++ b/drivers/iommu/iova.c @@ -746,8 +746,12 @@ int iova_domain_init_rcaches(struct iova_domain *iovad) =20 spin_lock_init(&cpu_rcache->lock); cpu_rcache->loaded =3D iova_magazine_alloc(GFP_KERNEL); + if (!cpu_rcache->loaded) { + ret =3D -ENOMEM; + goto out_err; + } cpu_rcache->prev =3D iova_magazine_alloc(GFP_KERNEL); - if (!cpu_rcache->loaded || !cpu_rcache->prev) { + if (!cpu_rcache->prev) { ret =3D -ENOMEM; goto out_err; } @@ -903,7 +907,11 @@ static void free_iova_rcaches(struct iova_domain *iova= d) break; for_each_possible_cpu(cpu) { cpu_rcache =3D per_cpu_ptr(rcache->cpu_rcaches, cpu); + if (!cpu_rcache->loaded) + break; iova_magazine_free(cpu_rcache->loaded); + if (!cpu_rcache->prev) + break; iova_magazine_free(cpu_rcache->prev); } free_percpu(rcache->cpu_rcaches); --=20 2.17.1 From nobody Sun Feb 8 18:31:33 2026 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4EED4EB64D9 for ; Tue, 11 Jul 2023 03:05:37 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231228AbjGKDFf (ORCPT ); Mon, 10 Jul 2023 23:05:35 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:48946 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231393AbjGKDFA (ORCPT ); Mon, 10 Jul 2023 23:05:00 -0400 Received: from szxga02-in.huawei.com (szxga02-in.huawei.com [45.249.212.188]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 4344410F6 for ; Mon, 10 Jul 2023 20:03:46 -0700 (PDT) Received: from dggpeml500022.china.huawei.com (unknown [172.30.72.55]) by szxga02-in.huawei.com (SkyGuard) with ESMTP id 4R0Qf11jyzzVhlh; Tue, 11 Jul 2023 11:02:33 +0800 (CST) Received: from huawei.com (10.175.112.208) by dggpeml500022.china.huawei.com (7.185.36.66) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.27; Tue, 11 Jul 2023 11:03:43 +0800 From: Zhang Zekun To: , , , , CC: , , , , Subject: [PATCH -next 2/2] iommu/iova: allocate iova_rcache->depot dynamicly Date: Tue, 11 Jul 2023 10:54:52 +0800 Message-ID: <20230711025452.35475-3-zhangzekun11@huawei.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20230711025452.35475-1-zhangzekun11@huawei.com> References: <20230711025452.35475-1-zhangzekun11@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.175.112.208] X-ClientProxiedBy: dggems701-chm.china.huawei.com (10.3.19.178) To dggpeml500022.china.huawei.com (7.185.36.66) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" In fio test with 4k,read,and allowed cpus to 0-255, we observe a performance decrease of IOPS. The normal IOPS can reach up to 1980k, but we can only get about 1600k. abormal IOPS: Jobs: 12 (f=3D12): [R(12)][99.3%][r=3D6220MiB/s][r=3D1592k IOPS][eta 00m:12= s] Jobs: 12 (f=3D12): [R(12)][99.4%][r=3D6215MiB/s][r=3D1591k IOPS][eta 00m:11= s] Jobs: 12 (f=3D12): [R(12)][99.4%][r=3D6335MiB/s][r=3D1622k IOPS][eta 00m:10= s] Jobs: 12 (f=3D12): [R(12)][99.5%][r=3D6194MiB/s][r=3D1586k IOPS][eta 00m:09= s] Jobs: 12 (f=3D12): [R(12)][99.6%][r=3D6173MiB/s][r=3D1580k IOPS][eta 00m:08= s] Jobs: 12 (f=3D12): [R(12)][99.6%][r=3D5984MiB/s][r=3D1532k IOPS][eta 00m:07= s] Jobs: 12 (f=3D12): [R(12)][99.7%][r=3D6374MiB/s][r=3D1632k IOPS][eta 00m:06= s] Jobs: 12 (f=3D12): [R(12)][99.7%][r=3D6343MiB/s][r=3D1624k IOPS][eta 00m:05= s] normal IOPS: Jobs: 12 (f=3D12): [R(12)][99.3%][r=3D7736MiB/s][r=3D1980k IOPS][eta 00m:12= s] Jobs: 12 (f=3D12): [R(12)][99.4%][r=3D7744MiB/s][r=3D1982k IOPS][eta 00m:11= s] Jobs: 12 (f=3D12): [R(12)][99.4%][r=3D7737MiB/s][r=3D1981k IOPS][eta 00m:10= s] Jobs: 12 (f=3D12): [R(12)][99.5%][r=3D7735MiB/s][r=3D1980k IOPS][eta 00m:09= s] Jobs: 12 (f=3D12): [R(12)][99.6%][r=3D7741MiB/s][r=3D1982k IOPS][eta 00m:08= s] Jobs: 12 (f=3D12): [R(12)][99.6%][r=3D7740MiB/s][r=3D1982k IOPS][eta 00m:07= s] Jobs: 12 (f=3D12): [R(12)][99.7%][r=3D7736MiB/s][r=3D1981k IOPS][eta 00m:06= s] Jobs: 12 (f=3D12): [R(12)][99.7%][r=3D7736MiB/s][r=3D1980k IOPS][eta 00m:05= s] The current struct of iova_rcache will have iova_cpu_rcache for every cpu, and these iova_cpu_rcaches use a common buffer iova_rcache->depot to exchange iovas among iova_cpu_rcaches. A machine with 256 cpus will have 256 iova_cpu_rcaches and 1 iova_rcache->depot per iova_domain. However, the max size of iova_rcache->depot is fixed to MAX_GLOBAL_MAGS which equals to 32, and can't grow with the number of cpus, and this can cause problem in some condition. Some drivers will only free iovas in their irq call back function. For the driver in this case it has 16 thread irqs to free iova, but these irq call back function will only free iovas on 16 certain cpus(cpu{0,16, 32...,240}). Thread irq which smp affinity is 0-15, will only free iova on cpu 0. However, the driver will alloc iova on all cpus(cpu{0-255}), cpus without free iova in local cpu_rcache need to get free iovas from iova_rcache->depot. The current size of iova_rcache->depot max size is 32, and it seems to be too small for 256 users (16 cpus will put iovas to iova_rcache->depot and 240 cpus will try to get iova from it). Set iova_rcache->depot grow with the num of possible cpus, and the decrease of IOPS disappear. Signed-off-by: Zhang Zekun --- drivers/iommu/iova.c | 26 ++++++++++++++++++++++---- 1 file changed, 22 insertions(+), 4 deletions(-) diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c index 3c784a28e9ed..df37a4501e98 100644 --- a/drivers/iommu/iova.c +++ b/drivers/iommu/iova.c @@ -238,6 +238,7 @@ static int __alloc_and_insert_iova_range(struct iova_do= main *iovad, =20 static struct kmem_cache *iova_cache; static unsigned int iova_cache_users; +static unsigned int max_global_mags; static DEFINE_MUTEX(iova_cache_mutex); =20 static struct iova *alloc_iova_mem(void) @@ -625,7 +626,6 @@ EXPORT_SYMBOL_GPL(reserve_iova); * will be wasted. */ #define IOVA_MAG_SIZE 127 -#define MAX_GLOBAL_MAGS 32 /* magazines per bin */ =20 struct iova_magazine { unsigned long size; @@ -641,7 +641,7 @@ struct iova_cpu_rcache { struct iova_rcache { spinlock_t lock; unsigned long depot_size; - struct iova_magazine *depot[MAX_GLOBAL_MAGS]; + struct iova_magazine **depot; struct iova_cpu_rcache __percpu *cpu_rcaches; }; =20 @@ -722,6 +722,13 @@ int iova_domain_init_rcaches(struct iova_domain *iovad) unsigned int cpu; int i, ret; =20 + /* + * the size of max global mags should growth with the num of + * cpus + */ + if (!max_global_mags) + max_global_mags =3D max_t(unsigned int, 32, num_possible_cpus()); + iovad->rcaches =3D kcalloc(IOVA_RANGE_CACHE_MAX_SIZE, sizeof(struct iova_rcache), GFP_KERNEL); @@ -733,6 +740,12 @@ int iova_domain_init_rcaches(struct iova_domain *iovad) struct iova_rcache *rcache; =20 rcache =3D &iovad->rcaches[i]; + rcache->depot =3D kcalloc(max_global_mags, sizeof(struct iova_magazine *= ), + GFP_KERNEL); + if (!rcache->depot) { + ret =3D -ENOMEM; + goto out_err; + } spin_lock_init(&rcache->lock); rcache->depot_size =3D 0; rcache->cpu_rcaches =3D __alloc_percpu(sizeof(*cpu_rcache), @@ -798,7 +811,7 @@ static bool __iova_rcache_insert(struct iova_domain *io= vad, =20 if (new_mag) { spin_lock(&rcache->lock); - if (rcache->depot_size < MAX_GLOBAL_MAGS) { + if (rcache->depot_size < max_global_mags) { rcache->depot[rcache->depot_size++] =3D cpu_rcache->loaded; } else { @@ -903,8 +916,12 @@ static void free_iova_rcaches(struct iova_domain *iova= d) =20 for (i =3D 0; i < IOVA_RANGE_CACHE_MAX_SIZE; ++i) { rcache =3D &iovad->rcaches[i]; - if (!rcache->cpu_rcaches) + if (!rcache->depot) + break; + if (!rcache->cpu_rcaches) { + kfree(rcache->depot); break; + } for_each_possible_cpu(cpu) { cpu_rcache =3D per_cpu_ptr(rcache->cpu_rcaches, cpu); if (!cpu_rcache->loaded) @@ -917,6 +934,7 @@ static void free_iova_rcaches(struct iova_domain *iovad) free_percpu(rcache->cpu_rcaches); for (j =3D 0; j < rcache->depot_size; ++j) iova_magazine_free(rcache->depot[j]); + kfree(rcache->depot); } =20 kfree(iovad->rcaches); --=20 2.17.1